Open source intelligence analysis method based on large language model

Abstract

The invention provides an open source intelligence analysis method based on a large language model. The open source intelligence analysis method comprises the steps that open source intelligence is collected based on multiple search engines; and analyzing open source intelligence based on a large language model. According to the method, background information crawling analysis is performed on an intelligence event by fully utilizing the ability of a large language model, and information of different retrieval sources is summarized and evaluated based on cue words to obtain a return result; a large language model scoring mechanism and a cyclic evaluation mechanism are innovated, and the illusion phenomena of political tendency and the like existing in information crawling are effectively avoided through a retrieval enhancement generation technology based on a large language model; the open source intelligence product generation quality is improved, and compared with a traditional manual search or ES retrieval mode, the flexibility is higher, and information collection is more accurate; in a real business scene, the generation time delay of a normalized open source intelligence product is about a minute level, and the method has high timeliness and practicability in engineering practice.

Description

Technical Field

[0001] The present invention relates to the technical field of open source intelligence analysis, and in particular to an open source intelligence analysis method based on a large language model. Background Art

[0002] Open Source Intelligence (OSINT) is the process of collecting, organizing and analyzing publicly available data and information to obtain intelligence. The data sources are wide-ranging, including the Internet, social media, government public information, academic papers, news reports, blogs, forums, etc. Its important value lies in the ability to quickly obtain a large amount of information at a relatively low cost, providing decision makers with comprehensive, accurate and timely intelligence support.

[0003] As a method of intelligence collection, processing and analysis based on public information or resources, open source intelligence analysis covers a wide range of technologies and skills, including information collection, language processing, data mining, network analysis, statistical modeling, etc. Intelligence analysts integrate information from different sources and use advanced tools and technologies to identify key information hidden in massive data, thereby supporting decision-making, risk assessment, threat intelligence and other activities. The goal of open source intelligence analysis is to improve the ability to predict intelligence events, mitigate potential risks, reveal emerging threats and enhance the basis for decision-making.

[0004] In today's era of information explosion, intelligence analysis faces an unprecedented amount of information. From the Internet, social media, news reports to academic literature, data sources are rich and diverse. These massive data contain information that is crucial to social, economic and national security. However, due to the huge amount of data, traditional manual analysis methods can no longer meet the needs of deep understanding and comprehensive grasp of data. The limited human resources are also one of the challenges facing current information processing. Even the most professional and efficient analysis team finds it difficult to quickly and comprehensively analyze information from all sources, which inevitably misses key details and limits the accurate assessment of events or threats. In such a situation, it is crucial to find an efficient and comprehensive information processing method. At present, although there are many methods for open source intelligence acquisition and analysis, in actual use, the most difficult problem is how to obtain comprehensive intelligence source information and how to perform in-depth data processing on hundreds of intelligence source information in a short time and summarize them into a comprehensive intelligence product. In short, the way to obtain open source data and the way to analyze and summarize are the bottlenecks of open source intelligence analysis.

[0005] Large Language Model (LLM) is an artificial intelligence model trained with a large amount of data. It has powerful natural language processing and data analysis capabilities. Its application in open source intelligence analysis is of great significance. It can quickly process massive amounts of open source information, extract valuable intelligence information, and greatly improve the efficiency of intelligence analysis. In addition, the large language model uses deep learning and data mining techniques to deeply explore the correlation and trends between information, which helps to discover key intelligence information hidden in massive amounts of information and provide decision makers with more comprehensive and in-depth intelligence support. Summary of the invention

[0006] In view of the problems existing in the above-mentioned traditional open source intelligence analysis method, the present invention provides an open source intelligence analysis method based on a large language model.

[0007] The present invention provides an open source intelligence analysis method based on a large language model, comprising the following steps:

[0008] Collect open source intelligence based on multiple search engines;

[0009] Analyzing open source intelligence based on large language models.

[0010] Furthermore, the collecting of open source intelligence based on multiple search engines includes the following sub-steps:

[0011] Expand search content;

[0012] Search and crawl data based on multiple search engines;

[0013] Conduct credibility assessment and content screening on search crawled information;

[0014] De-duplicate and summarize the screening information to obtain open source intelligence.

[0015] Furthermore, the expanded search content includes:

[0016] Based on the semantic understanding ability of the large language model, the relevant retrieval content of the required intelligence events is expanded, and one search content is expanded into multiple.

[0017] Furthermore, the data search and crawling based on multiple search engines includes:

[0018] Through the tool calling capabilities of the large language model, multiple search engines are called to perform search crawling.

[0019] Furthermore, the credibility assessment and content screening of the search crawled information includes:

[0020] By using the thinking chain technology and based on the logical reasoning ability of the large language model, the search crawling information returned by multiple search engines is grouped and evaluated to find logically inconsistent content. A voting mechanism is used to conduct credibility assessment and realize content screening.

[0021] Furthermore, the deduplication summary of the screening information includes:

[0022] Based on the analytical capabilities of the large language model, the evaluation and screening information is summarized and aggregated, and duplicate content is merged to form the final open source intelligence after deduplication.

[0023] Furthermore, the analyzing open source intelligence based on the large language model includes the following sub-steps:

[0024] Obtain basic intelligence information;

[0025] Analyze basic intelligence information;

[0026] Predict intelligence events based on intelligence information analysis results.

[0027] Furthermore, the basic information of obtaining intelligence includes:

[0028] Leverage the tool calling capabilities of the large language model to call multiple search engines to search for basic information related to intelligence topics from open source intelligence.

[0029] Based on the logical thinking ability of the large language model, the basic information of the search is classified and summarized, and based on the analytical ability of the large language model, the logically consistent result content is selected, and the possible logically inconsistent content is removed to obtain comprehensive basic intelligence information.

[0030] Furthermore, the basic information of the analysis intelligence includes:

[0031] Obtain possible key information about intelligence stakeholders through basic intelligence information;

[0032] Obtain as comprehensive a view of the relevant parties’ statements on the incident and possible interests through this key information;

[0033] Among them, the reflection mechanism of the large language model is adopted, and based on the analysis and evaluation of the analysis results by the large language model, modification suggestions are put forward, and the missing intelligence information is analyzed based on the modification suggestions, and the basic intelligence information acquisition step is returned again to supplement the basic intelligence information.

[0034] Furthermore, predicting intelligence events based on intelligence information analysis results includes:

[0035] Based on the results of the analysis of basic intelligence information, interest relations are grouped according to content to form a community of interests among the relevant parties. Based on the interest information, statements, gains and losses of the relevant parties regarding the incident, predictions of the incident are formed and open source intelligence products are generated to provide decision makers with comprehensive, accurate and timely intelligence support.

[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0037] 1. The present invention enhances the ability to analyze open source intelligence. By making full use of the natural language processing and semantic understanding capabilities of the large language model, it can efficiently process massive text data, identify hidden information in the text, and mine valuable intelligence content from it. It understands the context and objectives of the task through a task-driven learning method, improves the efficiency and accuracy of intelligence analysis, and better handles open source data with a wide range of information sources, fragmentation, and uneven quality, and improves the efficiency of open source intelligence acquisition and analysis.

[0038] 2. The present invention improves the efficiency of open source data retrieval. Based on the logical thinking ability of the large language model, the relevant retrieval content of the required intelligence data is expanded, and one search content is expanded into multiple based on semantics, and through the tool calling ability of the large language model, multiple search engines are called for search, and the advantages of multiple search engines are combined to reduce search bias, enhance the credibility of search results, improve the efficiency of open source intelligence retrieval and broaden the boundaries of information retrieval, which is conducive to extracting rich and valuable intelligence from complex open source data.

[0039] 3. The present invention innovates the information credibility assessment mechanism. Based on the logical thinking ability of the large language model, the results returned by multiple search engines are grouped and evaluated, and logically inconsistent content is found. A voting mechanism is used for cyclical evaluation until the compliance criteria are met. The analysis and return results are completed to achieve the credibility assessment of open source intelligence, and the retrieval enhanced generation (RAG) technology of the large language model is used to effectively avoid the problem of political tendency in crawled information.

[0040] 4. The present invention improves the interpretability of intelligence analysis. In the process of open source intelligence analysis, the logical correctness of information is analyzed based on the thinking chain technology, and background information of intelligence events is crawled and analyzed based on known facts or clues. Information from different search sources is analyzed and judged based on prompt words, and the missing intelligence information is further analyzed in combination with manual intervention suggestions, and the intelligence collection step is returned again to supplement the information, gradually building a complete intelligence context.

[0041] 5. The large language model in the present invention has powerful natural language processing and information understanding capabilities, and can efficiently process massive text data. Its deep semantic understanding ability enables it to identify hidden information in the text and mine valuable intelligence content from it. Through task-driven learning, the large language model can better understand the context and objectives of the task, and improve the efficiency and accuracy of intelligence analysis. Taking into account the characteristics of open source intelligence, this method can better handle situations where information sources are wide, fragmented, and of varying quality, and improve the efficiency of intelligence acquisition and analysis. Most importantly, this method has strong sustainability. As the large language model continues to learn and optimize in massive data, its performance in intelligence analysis will continue to improve, providing continuous support and assistance for intelligence work. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flowchart of an open source intelligence analysis method based on a large language model in an embodiment of the present invention.

[0043] Figure 2 The present invention is a flowchart of the open source intelligence acquisition in the open source intelligence analysis method based on the large language model in the embodiment of the present invention.

[0044] Figure 3 The present invention is a flowchart of open source intelligence analysis in an open source intelligence analysis method based on a large language model in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0046] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Example

[0048] like Figure 1 , Figure 2 , Figure 3 As shown, this embodiment proposes an open source intelligence analysis method based on a large language model, comprising the following steps:

[0049] Step 1: Collect open source intelligence based on multiple search engines:

[0050] In order to effectively mine and utilize massive, heterogeneous, and multi-source public information in order to reveal potential knowledge value and decision-making basis, if the information search coverage is narrow or insufficient in depth, it is easy to lead to the omission of key intelligence, affect the accuracy and completeness of the final analysis results, and even mislead the research direction. Figure 2 As shown, the open source intelligence collection based on the large language model specifically includes the following sub-steps:

[0051] Step 1.1, expand the search content. Based on the semantic understanding ability of the large language model, expand the relevant search content of the required intelligence events, and expand one search content into multiple based on the semantic analysis results.

[0052] Step 1.2, search and crawl data based on multiple search engines. Through the tool calling capability of the large language model, calling multiple search engines for search and crawling can integrate the advantages of multiple search engines, reduce search bias, enhance the credibility of search results, and improve the comprehensiveness of search.

[0053] Step 1.3, conduct credibility assessment and content screening of search crawled information. Using the mind chain technology, based on the logical reasoning ability of the large language model, the search crawled information returned by multiple search engines is grouped and evaluated, and logically inconsistent content is found. A voting mechanism is used to conduct credibility assessment and content screening.

[0054] Step 1.4: De-duplicate and summarize the evaluation and screening information. Based on the analytical capabilities of the large language model, the evaluation and screening information is summarized and aggregated, and duplicate content is merged to form the final de-duplicate open source intelligence.

[0055] Step 2: Analyze open source intelligence based on large language models:

[0056] First, multiple search engines are called based on the large language model to obtain the basic information of the required intelligence; second, the large language model is used to call the search engine based on the basic situation of the event to obtain the statements and possible interests of each party concerned about the event, and the interests are grouped according to the content to form a community of interests of the parties concerned; finally, based on the interest information, statements, gains and losses of the parties concerned about the event, a prediction of the event is formed. Figure 3 As shown, the open source intelligence analysis based on the large language model specifically includes the following sub-steps:

[0057] Step 2.1, obtain basic intelligence information:

[0058] First, the tool calling capability of the big language model is used to call search engines such as Google, DuckDuckGo, and Bing to search for basic information related to the intelligence topic from open source intelligence. Then, based on the logical thinking ability of the big language model, the basic searched information is classified and summarized, and based on the analytical ability of the big language model, the logically consistent result content is selected, and the possible logically inconsistent content is removed to obtain comprehensive basic intelligence information.

[0059] Step 2.2, analyze basic intelligence information:

[0060] Through the basic information of intelligence, we obtain possible key information about the parties related to the intelligence, such as location, country, etc., and then use this key information to obtain the most comprehensive statements of the parties related to the intelligence and possible interests. In the process of intelligence analysis, we adopt the thinking chain technology based on the logical thinking ability of the large language model to analyze the logical correctness of the information obtained, and based on the search of multiple search engines, ensure the political neutrality of the information. At the same time, we adopt the reflection mechanism of the large language model, based on the analysis and judgment of the analysis results of the large language model, put forward modification opinions, and analyze the missing intelligence information based on the modification opinions, and return to the basic information acquisition step to supplement the basic information of intelligence.

[0061] Step 2.3, predict intelligence events based on intelligence information analysis results:

[0062] Based on the results of the analysis of basic intelligence information, interest relations are grouped according to content to form a community of interests among the relevant parties. Based on the interest information, statements, gains and losses of the relevant parties regarding the incident, predictions of the incident are formed and open source intelligence products are generated to provide decision makers with comprehensive, accurate and timely intelligence support.

[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An open source intelligence analysis method based on a large language model, characterized in that: The steps include: Collect open source intelligence based on multiple search engines; Analyzing open source intelligence based on large language models.

2. The open source intelligence analysis method based on a large language model according to claim 1, characterized in that: The collecting of open source intelligence based on multiple search engines includes the following sub-steps: Expand search content; Data search and crawling based on multiple search engines; Conduct credibility assessment and content screening on search crawled information; De-duplicate and summarize the screening information to obtain open source intelligence.

3. The open source intelligence analysis method based on a large language model according to claim 2 is characterized in that: The expanded search content includes: Based on the semantic understanding ability of the large language model, the relevant retrieval content of the required intelligence events is expanded, and one search content is expanded into multiple.

4. The open source intelligence analysis method based on a large language model according to claim 2, characterized in that: The data search and crawling based on multiple search engines includes: Through the tool calling capabilities of the large language model, multiple search engines are called to perform search crawling.

5. The open source intelligence analysis method based on a large language model according to claim 2, characterized in that: The credibility assessment and content screening of search crawled information includes: By using the thinking chain technology and based on the logical reasoning ability of the large language model, the search crawling information returned by multiple search engines is grouped and evaluated to find logically inconsistent content. A voting mechanism is used to conduct credibility assessment and realize content screening.

6. The open source intelligence analysis method based on a large language model according to claim 1, characterized in that: The deduplication summary of the screening information includes: Based on the analytical capabilities of the large language model, the evaluation and screening information is summarized and aggregated, and duplicate content is merged to form the final open source intelligence after deduplication.

7. The open source intelligence analysis method based on a large language model according to claim 1, characterized in that: The analysis of open source intelligence based on a large language model includes the following sub-steps: Obtain basic intelligence information; Analyze basic intelligence information; Predict intelligence events based on intelligence information analysis results.

8. The open source intelligence analysis method based on a large language model according to claim 7, characterized in that: The basic information of obtaining intelligence includes: Leverage the tool calling capabilities of the large language model to call multiple search engines to search for basic information related to intelligence topics from open source intelligence. Based on the logical thinking ability of the large language model, the basic information of the search is classified and summarized, and based on the analytical ability of the large language model, the logically consistent result content is selected, and the possible logically inconsistent content is removed to obtain comprehensive basic intelligence information.

9. The open source intelligence analysis method based on a large language model according to claim 7, characterized in that: The basic information of the analytical intelligence includes: Obtain possible key information about intelligence stakeholders through basic intelligence information; Obtain as comprehensive a view of the relevant parties’ statements on the incident and possible interests through this key information; Among them, the reflection mechanism of the large language model is adopted, and based on the analysis and evaluation of the analysis results by the large language model, modification suggestions are put forward, and the missing intelligence information is analyzed based on the modification suggestions, and the basic intelligence information acquisition step is returned again to supplement the basic intelligence information.

10. The open source intelligence analysis method based on a large language model according to claim 7, characterized in that: The prediction of intelligence events based on intelligence information analysis results includes: Based on the results of the analysis of basic intelligence information, interest relations are grouped according to content to form a community of interests among the relevant parties. Based on the interest information, statements, gains and losses of the relevant parties regarding the incident, predictions of the incident are formed and open source intelligence products are generated to provide decision makers with comprehensive, accurate and timely intelligence support.

Images